This student's project aims to solve the problem of adverse drug reactions (ADRs), which is a significant cause of human death and morbidity worldwide, and costs the US healthcare system over $30 billion annually. Danika's solution combines drug chemical structures, human body reactions, and contextual information to predict ADRs by using a combination of deep learning and knowledge graph techniques and two datasets.

Danika used a novel pipeline built using Python, DeepChem, and PyKeen, and applied multi-task classification, graph convolutional neural networks, and knowledge graph embeddings. The multi-task classifier was found to be very effective in learning how different side effects are related to each other, and the graph convolutional networks were moderately effective. By combining information about drug side effects relationships, targets, and indications, the knowledge graph with TransE embedding was able to predict side effects effectively.

DrugGraph is a a holistic approach, combining drug chemistry, body interactions, and contextual information to improve the chances of predicting ADRs. Danika hopes to build an online solution that will be available to doctors, pharmacists, and patients worldwide, providing additional safety checks for drugs and reducing the harm of ADRs.